Conservative Systems and Hamiltonian Phase Flow

This chapter has been building a theory of flows $\dot{\mathbf{x}}=\mathbf{f}(\mathbf{x})$ that assume nothing: dissipation allowed, attractors expected, the Jacobian sorting equilibria into nodes, spirals, and saddles. Now meet the grand exception. When the flow conserves a quantity — an energy $E(\mathbf{x})$ exactly constant along every trajectory — the portrait is frozen before you solve anything: trajectories are pinned to the level curves of $E$ like hikers pinned to the contour lines of a map. One function of position, drawn as a contour plot, is the complete phase portrait. The consequences are drastic and beautifully negative. Nothing can spiral in: an attracting point would swallow an open basin of initial conditions, forcing $E$ constant across it — impossible for a genuine conserved quantity. So conservative systems have no attractors at all: no stable spirals, no stable nodes, no limit cycles. The only equilibria left are centers, ringed by families of closed orbits, and saddles, whose level curves cross to form separatrices. Better still, conservation rescues the one case where linearization fails — a borderline linear center becomes a certified nonlinear center the moment a conserved energy has a minimum there. Frictionless mechanics is where these systems come from, and the Hamiltonian articles — Hamiltonian Flow and the Phase Fluid, The Pendulum Phase Portrait and Separatrix — develop that structure in full. This article views the same systems through the nonlinear-dynamics lens: what conservation forbids, what it certifies, and what happens the instant you turn friction on.

Key takeaways

A flow $\dot{\mathbf{x}}=\mathbf{f}(\mathbf{x})$ is conservative when some nonconstant $E(\mathbf{x})$ has $\mathrm{d}E/\mathrm{d}t=0$ along every trajectory; orbits then lie on level sets of $E$ , so a contour plot of $E$ is the phase portrait.
Conservative systems have no attractors: an attracting point or cycle would force $E$ constant on an open basin, contradicting conservation. Nothing settles, nothing decays.
Equilibria of $\ddot{x}=-V'(x)/m$ sit at critical points of $V$ : minima are centers ( $\omega=\sqrt{V''/m}$ ), maxima saddles ( $\lambda=\pm\sqrt{-V''/m}$ ) — spirals and nodes structurally excluded.
The nonlinear centers theorem repairs linearization's blind spot: a nondegenerate minimum of $E$ at an isolated fixed point makes nearby orbits genuinely closed, even where the eigenvalue test is inconclusive.
Periods come from energy alone: $T(E)=\sqrt{2m}\int \mathrm{d}q/\sqrt{E-V(q)}$ — amplitude-independent only for the harmonic well, divergent at a separatrix, and increasing as amplitude shrinks in a quartic well.
The slightest damping gives $\mathrm{d}E/\mathrm{d}t=-b\dot{x}^2\le 0$ : the conserved quantity degrades into a Lyapunov function, orbit families collapse, and centers become the spirals of the rest of this chapter.
Key idea: A conserved energy freezes the phase portrait onto the level curves of $E$ — attractors are impossible, minima of $E$ are certified nonlinear centers, saddles of $E$ organize the separatrices, and the entire dissipative world of this chapter begins the moment that conservation is broken.

The core insight

A conserved energy pins every trajectory to a level curve of $E(\mathbf{x})$ , which forbids attractors outright, certifies minima of $E$ as true nonlinear centers, and leaves centers and saddles as the only possible equilibria — the structural opposite of a dissipative flow.

Where this lives today

The no-attractor theorem is now an engineering specification. Long-horizon orbital mechanics — JPL ephemerides, the billion-year Solar System integrations behind statements like "Mercury has a ~1% chance of destabilizing" — runs on symplectic integrators: schemes built to inherit the conservative structure exactly, because an ordinary integrator's tiny artificial dissipation manufactures exactly the spurious spirals this article proves cannot exist, and after $10^{12}$ steps the lie would dominate the answer. Accelerator physicists live by the same discipline: a stored LHC bunch completes hundreds of millions of turns per fill, and tracking codes must keep the numerical flow attractor-free or predict beam losses that are pure artifact. The 2026 twist is that the structure is being learned. Hamiltonian neural networks and their symplectic descendants fit a conserved $H$ — rather than a raw vector field — to data from molecular dynamics, plasma, and robotics, hard-wiring the no-attractor guarantee into the model class: an architecture whose inductive bias is literally Derivation A. In the laboratory, trapped-ion and optomechanics groups now reconstruct phase-space contours of nearly conservative mesoscopic oscillators directly, watching a center's neutral rings shade into the faint spirals of residual dissipation — the hero figure's two panels, measured in one apparatus.

The physical picture

Take the double-well system $\ddot{x}=-V'(x)/m$ with $V(x)=2x^4-4x^2$ — a particle free to sit in either of two valleys separated by a hump. Without friction, energy conservation dictates the entire $(x,\dot{x})$ portrait: two centers at the valley bottoms ringed by nested closed orbits, a saddle on the hump, and through the saddle a figure-eight separatrix dividing "trapped in one well" from "swinging across both." Every curve is a contour of one function, $E=\tfrac12 m\dot{x}^2+V(x)$ . Now add a whisper of friction. The skeleton of equilibria survives, but the flesh changes completely: the nested families collapse, every trajectory cuts across the old contours, bleeding downhill in energy, and the two centers become spiral attractors competing for territory. The saddle's stable manifold — formerly the separatrix — becomes the boundary between two basins. Same landscape, opposite dynamics: this pair of portraits is the chapter's clearest statement of what dissipation does.

Figure

Same double well with and without friction — conservation pins orbits to energy contours, dissipation lets them bleed across the contours into the two spiral attractors

Why is "no attractors" so automatic? Because attraction is a statement about open sets of initial conditions, and conservation defeats it. A would-be stable spiral must visit contours of ever-lower $E$ , but conservation forbids leaving the contour it started on. This is not an eigenvalue computation — it is a one-line topological veto, and it applies to limit cycles as much as to points.

Figure

Why nothing can spiral in — a trajectory of a conservative flow rides a single energy contour while an attracting spiral would have to cut across the whole nest of contours

The same law that forbids attraction also rescues the borderline case of the last two articles. Linear stability analysis warned that a pure-imaginary eigenvalue pair is fragile evidence — cubic terms can nudge a linear center into a slow spiral, and the Jacobian cannot tell. But a spiral, however slow, changes $E$ . If the conserved energy has a genuine minimum at the fixed point, its nearby level sets are closed rings, the trajectory is trapped on one ring, and the orbit must close. A second, subtler symmetry produces center-filled portraits without any conserved quantity: reversibility. A system invariant under $t\to-t$ , $y\to-y$ (a frictionless mechanism watched on rewound film) pairs every trajectory with a mirror twin, and the pairing forces orbits that cross the symmetry axis twice to close. Conservative and reversible overlap heavily in mechanics, but neither implies the other — and both, independently, veto the generic dissipative behavior this chapter otherwise expects.

Figure

Reversibility pairs every trajectory with a mirror twin — an arc above the axis and its reflection below, traversed in the opposite sense, joining into a closed orbit

What conservation means for a flow

The nonlinear-dynamics definition is deliberately broader than mechanics. A system $\dot{\mathbf{x}}=\mathbf{f}(\mathbf{x})$ on the plane is conservative when some scalar function is frozen by the flow — and to exclude trivial choices (a constant function is "conserved" by anything), we demand $E$ be nonconstant on every open set:

\frac{\mathrm{d}E}{\mathrm{d}t}=\nabla E\cdot\mathbf{f}(\mathbf{x})=0 \quad\text{along trajectories} \qquad (1)

where: $E(\mathbf{x})$ — the conserved quantity (energy, in mechanical examples); $\nabla E$ — its gradient, everywhere perpendicular to the flow. Geometrically, (1) says $\mathbf{f}$ is tangent to the level sets of $E$ : the flow field and the contour map are the same picture. Any trajectory that starts on the curve $E=E_0$ stays on it forever. The mechanical workhorse is a particle in a one-dimensional potential, with Newton's law recast as a planar flow:

\dot{x}=y,\qquad \dot{y}=-\frac{V'(x)}{m},\qquad E=\tfrac12 my^{2}+V(x) \qquad (2)

where: $y$ — the velocity treated as an independent coordinate; $V(x)$ — the potential. Check conservation directly: $\dot{E}=my\dot{y}+V'\dot{x}=y(-V')+V'y=0$ . Every frictionless mass-and-potential problem you have ever solved is a conservative planar flow, and its phase portrait is the contour plot of (2). For the Hamiltonian formulation — the same physics with $(q,p)$ coordinates and far-reaching consequences — one line suffices here:

\dot{q}=\frac{\partial H}{\partial p},\qquad \dot{p}=-\frac{\partial H}{\partial q},\qquad \frac{\mathrm{d}H}{\mathrm{d}t}=\frac{\partial H}{\partial t} \qquad (3)

where: $H(q,p)$ — the Hamiltonian, conserved whenever it lacks explicit time dependence (the two flow terms cancel by the chain rule). The article Hamiltonian Flow and the Phase Fluid develops what is special beyond conservation — the flow field $(\partial H/\partial p,\,-\partial H/\partial q)$ is exactly divergence-free, making the flow area-preserving (Liouville's theorem) — and Fixed Points and Stability in Phase Space derives the traceless-Jacobian consequence in full. Here we need only the punchline of that analysis, because it slots the conservative world into this chapter's classification machinery:

\operatorname{tr}J=0 \;\Longrightarrow\; \lambda_{1,2}=\pm\sqrt{-\det J\,},\qquad \det J\Big|_{\text{eq}}=\frac{V''(q_0)}{m} \qquad (4)

where: $J$ — the Jacobian at an equilibrium $(q_0,0)$ with $V'(q_0)=0$ . On the trace–determinant plane of the next article, conservative equilibria live pinned to the vertical axis $\tau=0$ : eigenvalues come as $\pm i\omega$ (a center, when $V''>0$ ) or $\pm\lambda$ (a saddle, when $V''<0$ ). The whole interior of the classification diagram — nodes, spirals, everything with $\tau\neq0$ — is structurally off-limits. Bounded orbits ring the centers, and their clock is set by energy alone. Solving (2) for $y=\dot{x}$ and separating variables converts the closed orbit into an integral between its turning points $q_\pm$ (where $V(q_\pm)=E$ ):

T(E)=\sqrt{2m}\,\int_{q_-}^{q_+}\frac{\mathrm{d}q}{\sqrt{E-V(q)}} \qquad (5)

where: $T$ — the period of the orbit with energy $E$ . Three regimes repay memorizing: a harmonic well gives the amplitude-independent $T=2\pi\sqrt{m/V''}$ ; a generic anharmonic well makes $T$ depend on $E$ (the double well's orbits slow as they approach the separatrix, where $T\to\infty$ logarithmically); and a flat-bottomed quartic well produces the genuinely counterintuitive $T\propto E^{-1/4}$ — smaller swings take longer — worked to numbers in Example 2. Finally, the contrast that frames this whole article. Add linear drag to (2) and recompute the energy's fate:

m\ddot{x}=-V'(x)-b\dot{x} \;\Longrightarrow\; \frac{\mathrm{d}E}{\mathrm{d}t}=-b\,\dot{x}^{2}\;\le\;0 \qquad (6)

where: $b$ — the drag coefficient. $E$ is no longer conserved but monotone: a Lyapunov function, strictly decreasing except at instants when $\dot{x}=0$ . Trajectories now cut across the old contours, always downhill, and must terminate on the set of equilibria — Derivation C makes this precise. One sign flip in one term, and every theorem of this article evaporates.

Which tool answers which question

You want to know	Reach for	Because
the entire phase portrait at once	contour plot of $E(x,y)$ from (2)	trajectories are the level curves
whether an equilibrium oscillates or escapes	sign of $V''(q_0)$ via (4)	minima are centers, maxima saddles; nothing else exists
whether a linear center is trustworthy	nonlinear centers theorem (Derivation B)	conservation closes the gap linearization leaves
whether anything can settle down	no-attractor theorem (Derivation A)	conserved $E$ on an open basin is a contradiction
the oscillation period at energy $E$	quadrature (5)	one integral, no solving of the ODE
what damping will do	$\dot{E}=-b\dot{x}^2$ from (6)	$E$ becomes a Lyapunov function; centers become spirals

Derivation AWhy a conservative system cannot have an attractor

We prove the signature theorem of this article: a flow with a conserved quantity $E$ — nonconstant on every open set — admits no attracting fixed point, and by the same argument no attracting closed orbit. The strategy is topological rather than computational: we let the supposed attractor act on its own basin of attraction, and watch continuity of $E$ produce a contradiction. Watch how little is used — no eigenvalues, no explicit form of $\mathbf{f}$ , only the definition of attraction and the continuity of $E$ .

Step 1. Suppose $\mathbf{x}^*$ is an attracting fixed point. By definition there is an open set $U$ — its basin, or any open ball inside the basin — such that every trajectory starting in $U$ satisfies $\mathbf{x}(t)\to\mathbf{x}^*$ as $t\to\infty$ .

Step 2. Evaluate $E$ along one such trajectory. Conservation makes it constant in time: $E(\mathbf{x}(t))=E(\mathbf{x}(0))$ for all $t$ .

Step 3. Take the limit using continuity of $E$ :

E(\mathbf{x}(0))=\lim_{t\to\infty}E(\mathbf{x}(t))=E\!\left(\lim_{t\to\infty}\mathbf{x}(t)\right)=E(\mathbf{x}^*)

Why continuity is the load-bearing hypothesis: the limit may be passed inside $E$ only because $E$ is continuous — this is the single analytic fact the proof consumes.

Step 4. The starting point $\mathbf{x}(0)$ was arbitrary in $U$ , so $E\equiv E(\mathbf{x}^*)$ on the entire open set $U$ . But a conserved quantity was required to be nonconstant on every open set — contradiction. No attracting fixed point exists.

Step 5. Repelling fixed points fall to the same argument run in reverse time ( $t\to-t$ preserves conservation), and an attracting limit cycle fails identically: its basin is an open annulus on which $E$ would again be forced constant. Conservative systems have no attractors and no repellers of any kind — only neutral structures. Why this changes what "stability" can mean here: the strongest stability available in a conservative system is Lyapunov (neighbors stay near) — asymptotic stability (neighbors converge) is off the table globally. *Verified: Strogatz's Theorem 6.5.1 argument; consistency check — the damped system (6), which does have attractors, indeed has no conserved $E$ : the candidate decays as $\dot{E}=-b\dot{x}^2<0$ , so the hypothesis fails exactly where the conclusion does.*

Go deeper · Derivation B · 2 minThe nonlinear centers theorem — when a linear center is the real thing

Derivation BThe nonlinear centers theorem — when a linear center is the real thing

Linearization at a center is inconclusive: eigenvalues $\pm i\omega$ sit on the knife-edge $\operatorname{tr}J=0$ , and unmodeled cubic terms can tip a general flow into a slow spiral. We now prove that a conserved energy with a local minimum forbids the tipping: if $\mathbf{x}^*$ is an isolated fixed point of a conservative system and a nondegenerate local minimum of $E$ , then every trajectory sufficiently close to $\mathbf{x}^*$ is a closed orbit. The strategy is to trade the differential equation for the geometry of level sets near a minimum, then use uniqueness of solutions to make the trajectory traverse its whole level curve.

Step 1. Near a nondegenerate minimum, $E$ looks like a bowl. Expanding, $E(\mathbf{x})-E(\mathbf{x}^*)=\tfrac12\,\delta\mathbf{x}^{\!\top} \mathbf{H}_E\,\delta\mathbf{x}+O(|\delta\mathbf{x}|^3)$ with positive-definite Hessian $\mathbf{H}_E$ , so for small $\epsilon>0$ the level set $E=E(\mathbf{x}^*)+\epsilon$ is a single closed curve encircling $\mathbf{x}^*$ — a slightly distorted ellipse. Why nondegeneracy matters at this step: a positive-definite Hessian is what guarantees the contours are closed rings; for a degenerate minimum (like the quartic well of Example 2, where the Hessian direction along $x$ vanishes) the same conclusion holds but must be checked from the exact $E$ , since higher-order terms shape the contour.

Step 2. Trajectories live on level sets by (1). A trajectory starting on the ring $E=E(\mathbf{x}^*)+\epsilon$ remains on that one-dimensional closed curve for all time.

Step 3. The trajectory cannot stall. Flow speed vanishes only at fixed points; $\mathbf{x}^*$ is isolated, so for small enough $\epsilon$ the ring contains no fixed point and $|\mathbf{f}|$ attains a positive minimum on it (a continuous positive function on a compact curve).

Step 4. A point moving one way along a closed curve of finite length, at speed bounded below, returns to its start in finite time $T\le(\text{length})/(\min|\mathbf{f}|)$ ; by uniqueness the motion thereafter repeats. Every small-enough ring is one closed orbit, so $\mathbf{x}^*$ is a true nonlinear center.

Step 5. Note what was not assumed: no smallness of nonlinearity, no Hamiltonian structure — only a conserved $E$ with a minimum. The theorem thus certifies the pendulum's bottom equilibrium, the double well's valleys, and even the degenerate quartic bottom, where the eigenvalue test gives up entirely. Verified: Strogatz Theorem 6.5.2 (with the traversal argument spelled out); cross-check against (4) — at a nondegenerate minimum $V''>0$ gives eigenvalues $\pm i\sqrt{V''/m}$ , and the theorem upgrades "linear center" to "nonlinear center", matching the closed contours of $E=\tfrac12 my^2+V$ directly.

Go deeper · Derivation C · 2 minHow dissipation dismantles the conservative portrait

Derivation CHow dissipation dismantles the conservative portrait

We add linear drag to the conservative particle and track exactly which structures survive. The goal is the precise statement behind the hero figure: with $b>0$ , the old energy becomes a strict Lyapunov function, all bounded trajectories terminate on equilibria, minima of $V$ turn from centers into asymptotically stable spirals (or nodes), and saddles remain saddles whose stable manifolds become basin boundaries. The strategy: compute $\dot{E}$ once, then squeeze the limit set with a LaSalle-type argument, then reclassify with the Jacobian trace.

Step 1. Energy budget. For $m\ddot{x}=-V'(x)-b\dot{x}$ , with $E=\tfrac12 m\dot{x}^2+V(x)$ :

\frac{\mathrm{d}E}{\mathrm{d}t}=\dot{x}\,(m\ddot{x}+V'(x))=\dot{x}\,(-b\dot{x})=-b\,\dot{x}^{2}\;\le\;0

Why this single line already forbids closed orbits: around any nonequilibrium periodic orbit, $\oint\dot{E}\,\mathrm{d}t=-b\oint\dot{x}^2\mathrm{d}t<0$ , yet $E$ must return to its starting value on a closed loop — contradiction. The entire one-parameter family of conservative orbits dies at once, for any $b>0$ .

Step 2. Where trajectories can end. On a bounded trajectory $E(t)$ decreases and is bounded below, hence converges, so $\dot{E}\to0$ and thus $\dot{x}\to0$ ; persisting on $\dot{x}=0$ requires $V'(x)=0$ too. The limit set consists of equilibria: every bounded motion funnels into a critical point of $V$ .

Step 3. Reclassify the minima. The Jacobian at $(q_0,0)$ now reads

J=\begin{pmatrix}0&1\\-V''(q_0)/m&-b/m\end{pmatrix},\qquad \operatorname{tr}J=-\frac{b}{m}<0,\qquad \det J=\frac{V''(q_0)}{m}

At a minimum ( $\det J>0$ ) the negative trace lands the equilibrium in the stable region of the trace–determinant plane: a stable spiral for weak drag ( $b^2<4mV''$ , eigenvalues $-\tfrac{b}{2m}\pm i\omega'$ with $\omega'=\sqrt{V''/m-b^2/4m^2}$ ), a stable node for strong drag. Lyapunov stability has been upgraded to asymptotic stability — the upgrade conservation forbade.

Step 4. Saddles survive in character: $\det J=V''/m<0$ keeps the eigenvalues real and opposite-signed for any $b$ . But their role changes — with attractors now existing, the saddle's stable manifold becomes the basin boundary between initial conditions falling into one well or the other (the hero figure, right panel).

Step 5. Numbers for the double well of Example 1 ( $m=0.5$ kg, $V''=16$ J m $^{-2}$ at the minima) with mild drag $b=0.1$ kg s $^{-1}$ : decay rate $b/2m=0.1$ s $^{-1}$ , ringing frequency $\omega'=\sqrt{32-0.01}\approx5.66$ rad s $^{-1}$ — indistinguishable from the undamped $5.657$ . The orbit looks conservative on any single swing yet loses an e-fold of amplitude every $10$ s, about nine oscillations. Weak dissipation hides in plain sight; the topology has nonetheless completely changed. Verified: $\dot{E}=-b\dot{x}^2$ recomputed by direct substitution; eigenvalues from $\lambda^2+(b/m)\lambda+V''/m=0$ give $-0.1\pm5.656i$ for the quoted numbers, matching $\omega'^2=32-0.01$ ; consistent with the damped-oscillator classification of Strogatz §5.2 and the Lyapunov/LaSalle treatment in standard texts.

Limiting cases

Case	Result	Meaning
$V''(q_0)>0$ (minimum)	center; $\omega=\sqrt{V''/m}$ , small orbits closed	neutral oscillation certified by the nonlinear centers theorem — Lyapunov stable, never asymptotically
$V''(q_0)<0$ (maximum)	saddle; $\lambda=\pm\sqrt{-V''/m}$	exponential flight along the unstable manifold; level curve through it is a separatrix
$V''(q_0)=0$ , quartic bottom	linearization silent (double zero eigenvalue)	energy contours still close: a degenerate center; $T\propto A^{-1}$ , diverging as $A\to0$ (Example 2)
$E\to E_{\text{sep}}$	$T(E)\to\infty$ logarithmically	the orbit lingers near the saddle for a time $\sim\lambda^{-1}\ln(1/\|E-E_{\text{sep}}\|)$
$b\to0^{+}$	attractors appear for any $b>0$ ; family of closed orbits vanishes	conservation is structurally unstable — the portraits of this article are the boundary case of the chapter
$E$ far above all wells	orbit encircles the whole potential landscape	contour approaches $y\approx\pm\sqrt{2E/m}$ : nearly free motion, gently modulated by $V$

Common trapThe orbits close, so the system must be settling into a rhythm

It is tempting to read a center's closed orbits as a destination — as if the system "wants" to end up on them the way a damped-driven system ends up on a limit cycle. It cannot. A center is only neutrally stable: displace the state and it adopts a different closed orbit at the new energy, with a different amplitude and (generically) period, forever remembering the perturbation. Nothing relaxes, because Derivation A forbids anything in a conservative flow from attracting anything. Limit Cycles and Self-Sustained Oscillations studies exactly the opposite — an isolated orbit that neighbors converge onto — and it will require breaking conservation. Related trap: do not "verify" a center with an off-the-shelf integrator; a non-symplectic scheme injects artificial dissipation, and its spurious slow spiral is the very artifact this article proves cannot be real.

Worked examples

1 · Full census of a double well (confidence builder)

A particle of mass $m=0.5$ kg moves in $V(x)=2x^4-4x^2$ (joules, $x$ in metres). Find and classify every equilibrium, give the oscillation frequency at the stable ones, the separatrix energy, and the fate of a particle launched from $x=1$ m with speed $2$ m s $^{-1}$ , then with $3$ m s $^{-1}$ .

Step 1. Equilibria from $V'(x)=8x^3-8x=8x(x^2-1)=0$ : three of them, $x=0$ and $x=\pm1$ (all with $\dot{x}=0$ ).

Step 2. Classify with $V''(x)=24x^2-8$ . At $x=0$ : $V''=-8$ J m $^{-2}<0$ — a potential maximum, hence by (4) a saddle with $\lambda=\sqrt{-V''/m}=\sqrt{8/0.5}=4.0$ s $^{-1}$ . At $x=\pm1$ : $V''=16$ J m $^{-2}>0$ — minima, hence centers with $\omega=\sqrt{16/0.5}=\sqrt{32}=5.66$ rad s $^{-1}$ , small-oscillation period $T=2\pi/\omega=1.11$ s.

Step 3. Separatrix energy: the level of $E$ through the saddle, $E_{\text{sep}}=V(0)=0$ . Well depth: $V(\pm1)=2-4=-2$ J. Orbits with $-2<E<0$ are trapped in one well; orbits with $E>0$ swing across both.

Step 4. Launch 1: $E=\tfrac12(0.5)(2^2)+V(1)=1.0-2.0=-1.0$ J $<0$ — trapped in the right well. Turning points from $2x^4-4x^2=-1$ : $x^2=1\pm\tfrac{\sqrt2}{2}$ , giving $x=0.541$ m and $x=1.307$ m. The particle shuttles between them, never reaching the hump.

Step 5. Launch 2: $E=\tfrac12(0.5)(3^2)-2.0=2.25-2.0=+0.25$ J $>0$ — above the separatrix. It crosses the hump at speed $v$ with $\tfrac12(0.5)v^2=0.25$ , i.e. $v=1.0$ m s $^{-1}$ , and sweeps the full landscape out to turning points $x^2=[2+\sqrt{4.5}]/2=2.061$ , $x=\pm1.436$ m. One curve of the contour map decided everything; no differential equation was solved.

2 · The quartic well that swings slower the smaller the swing (exam level)

A particle of mass $m=1$ kg sits in the flat-bottomed well $V(x)=\alpha x^4$ with $\alpha=1$ J m $^{-4}$ . Linearization at the bottom fails completely ( $V''(0)=0$ ). Show the origin is nonetheless a center, find $T(A)$ for amplitude $A$ , and evaluate it at $A=0.5$ , $1$ , and $2$ m.

Step 1. The energy $E=\tfrac12\dot{x}^2+x^4$ has a strict (though degenerate) minimum at the origin, and its level sets $\tfrac12 y^2+x^4=E$ are closed, rounded-rectangle curves for every $E>0$ . By the nonlinear centers theorem (Derivation B, with the degenerate-minimum caveat checked directly on these contours), every orbit near the origin is closed: a genuine center that the Jacobian — eigenvalues both zero — could never certify.

Step 2. Period from the quadrature (5) with $E=\alpha A^4$ and turning points $\pm A$ :

T=4\int_0^A\frac{\mathrm{d}x}{\sqrt{(2\alpha/m)(A^4-x^4)}}=\frac{4}{A}\sqrt{\frac{m}{2\alpha}}\int_0^1\frac{\mathrm{d}u}{\sqrt{1-u^4}}

using the substitution $x=Au$ — the amplitude scales out as $1/A$ .

Step 3. The remaining number is a Beta-function value: $\int_0^1(1-u^4)^{-1/2}\mathrm{d}u=\tfrac14 B(\tfrac14,\tfrac12)=\frac{\Gamma(1/4)\,\Gamma(1/2)}{4\,\Gamma(3/4)}=1.3110$ .

Step 4. Assemble: $T(A)=\dfrac{4\times1.3110}{A}\sqrt{\dfrac{1}{2}}=\dfrac{3.708}{A}$ seconds ( $A$ in metres). So $T(0.5)=7.42$ s, $T(1)=3.71$ s, $T(2)=1.85$ s. Halving the amplitude doubles the period.

Step 5. Read it physically: near the flat bottom the restoring force $-4\alpha x^3$ is feeble, so a small-amplitude particle dawdles; a large-amplitude one is yanked back hard and hurries. Dimensional analysis fixes the scaling — $v\sim\sqrt{E}\sim A^2$ while distance per cycle $\sim A$ , so $T\sim A/v\sim1/A$ . Isochronism is a property of harmonicity, not of conservation.

The physicist

Origins · 1 minWilliam Rowan Hamilton and how this was discovered

William Rowan Hamilton

1805–1865 · Dublin; Trinity College and Dunsink Observatory

Hamilton reorganized mechanics twice before turning thirty — first with a characteristic function borrowed from his optics, then, in the great 1834–35 papers, with the pair of first-order equations that bear his name. What matters for this article is what his formulation silently guarantees: write any frictionless system in Hamiltonian form and conservation of $H$ , the traceless Jacobian, and the center-and-saddle-only portrait follow automatically — the structural skeleton nonlinear dynamics would learn to see, a half-century later, through Poincaré's geometric eyes. Hamilton himself never drew a phase portrait; he was chasing an optical-mechanical unification that looked almost mystical to contemporaries and finally paid off in 1926, when Schrödinger used it to pass from classical trajectories to wave mechanics. The dissipative world beyond his equations — where this chapter now heads — is everything his framework was designed to exclude.

History

1687

— Newton's Principia poses the dynamics of particles under forces; energy is implicit, not yet an organizing principle

1788

— Lagrange's Mécanique analytique recasts mechanics variationally, the platform Hamilton will build on

1834

— Hamilton publishes the canonical equations; conservative dynamics becomes the flow of a single function on phase space

1838

— Liouville proves the phase-volume conservation theorem that makes Hamiltonian flow incompressible

1881

— Poincaré begins the qualitative theory of differential equations: portraits, not formulas, as the object of study

1892

— Lyapunov's stability memoir separates neutral from asymptotic stability — the distinction conservation makes unavoidable

1954

— Kolmogorov announces the persistence of quasi-periodic orbits under small perturbations, seeding KAM theory

Going further

References. Strogatz, Nonlinear Dynamics and Chaos, 2nd ed., §6.5–6.6 (conservative systems, reversibility) · Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., ch. 8–9 (Hamilton's equations, Liouville's theorem) · Landau & Lifshitz, Mechanics, 3rd ed., §§6–7, 11 (energy conservation, phase trajectories, the period integral) · Taylor, Classical Mechanics, ch. 13 (Hamiltonian mechanics) · Morin, Introduction to Classical Mechanics (oscillations, the pendulum)

Provenance. Equations are standard Hamiltonian mechanics and Strogatz §6.5–6.6 nonlinear dynamics. A verifier should recompute: (1) $\dot{E}=0$ for system (2) and $\mathrm{d}H/\mathrm{d}t=\partial H/\partial t$ from Hamilton's equations (Goldstein ch. 8); (2) the traceless-Jacobian eigenvalues $\pm\sqrt{-\det J}$ with $\det J=V''/m$ (Strogatz §6.5); (3) the no-attractor and nonlinear-centers arguments (Strogatz Theorems 6.5.1–6.5.2); (4) the double-well census — equilibria $0,\pm1$ ; $\lambda=4.0$ s $^{-1}$ ; $\omega=5.66$ rad s $^{-1}$ ; $E_{\text{sep}}=0$ ; turning points $0.541/1.307$ m at $E=-1$ J and $\pm1.436$ m at $E=+0.25$ J; (5) the quartic period $T=(3.708\,\text{s}\,\cdot\,\text{m})/A$ via $\int_0^1(1-u^4)^{-1/2}\mathrm{d}u=\Gamma(\tfrac14)\Gamma(\tfrac12)/4\Gamma(\tfrac34)=1.3110$ (Landau §11); (6) the damped reclassification, eigenvalues $-b/2m\pm i\omega'$ with the quoted numbers. No fabricated constants.

Scope note. This article is the nonlinear-dynamics view of conservative flow. The Hamiltonian formalism itself — canonical structure, Liouville's theorem and the phase fluid, the pendulum's exact period and separatrix, the oscillator's rotating ellipse — lives in Hamiltonian Flow and the Phase Fluid, The Pendulum Phase Portrait and Separatrix, and The Harmonic Oscillator in Phase Space.

Builds on: Phase Portraits and the Phase Plane · Fixed Points and Equilibria · Linear Stability Analysis and the Jacobian Leads to: Classification of Fixed Points: Nodes, Saddles, Spirals, Centers · Limit Cycles and Self-Sustained Oscillations · Hopf Bifurcation and Birth of Limit Cycles

Is "conservative" the same thing as "Hamiltonian"?

Not quite — conservative is the broader notion. Conservative asks only for some conserved function $E$ ; Hamiltonian demands the flow be generated by it through $\dot{q}=\partial H/\partial p$ , $\dot{p}=-\partial H/\partial q$ , which additionally makes the flow exactly area-preserving (the phase-fluid story of Hamiltonian Flow and the Phase Fluid). Every autonomous 1-DOF Hamiltonian system is conservative; the converse fails — rescale time by a positive function of position and you keep the same orbits (same conserved $E$ , same portrait) while wrecking area-preservation. For this article's qualitative questions — attractors, centers, saddles — the conserved function alone does all the work.

If nothing in a conservative system ever settles down, how does anything in the real world come to rest?

Because the real world is never exactly conservative — and the punchline is that the difference, however small, is qualitative. Derivation C shows any $b>0$ converts the conserved energy into a strictly decreasing Lyapunov function, so rest states exist for real pendulums and real molecules. What the idealization buys you is the correct intermediate-time picture (an LHC proton over $10^8$ turns, the Earth–Moon system over megayears) plus the structural skeleton — equilibria, their center/saddle types, separatrices — which weak damping deforms but does not relocate. The art is knowing which clock you are on: shorter than $m/b$ , trust the contours; longer, trust the spiral.

The no-attractor proof looks too easy — where would it break for a damped system, exactly?

At the hypothesis, immediately. For $m\ddot{x}=-V'-b\dot{x}$ , the natural candidate $E=\tfrac12 m\dot{x}^2+V$ obeys $\dot{E}=-b\dot{x}^2<0$ whenever the particle moves, so $E$ is not conserved and Step 2 of Derivation A never starts. Could some other clever function be conserved instead? No — run the argument in reverse: the damped double well demonstrably has attracting spirals, and Derivation A shows any continuous conserved quantity would be forced constant on their basins, hence on open sets, hence illegitimate. Attractors and conservation laws are logically exclusive; each one's existence proves the other's absence.

Why do people say a center is "structurally unstable" when the nonlinear centers theorem just proved it is robust?

Two different perturbations are in play. The theorem perturbs the initial condition within a fixed conservative system: nearby starts give nearby closed orbits — Lyapunov stability, rock solid. Structural stability perturbs the equations: add a tiny dissipative term to $\mathbf{f}$ and the center becomes a spiral — the topology changes, so the center is structurally unstable among generic flows. Resolution: within the class of conservative systems the center is robust (perturb $V$ and minima persist, with their centers); against perturbations that leave the class — friction, driving — it is defenseless. Knowing which class your perturbation respects is the whole game.

Where does the logarithmic divergence of the period near a separatrix actually come from?

From the exponential slowdown at the saddle. Near a saddle the linearization rules: distance along the incoming direction shrinks like $e^{-\lambda t}$ , with $\lambda=\sqrt{-V''/m}$ . An orbit at energy $E$ just below the separatrix creeps past the saddle at closest approach $\propto\sqrt{|E-E_{\text{sep}}|}$ ; entering from and leaving to fixed distances therefore takes $\approx\tfrac{2}{\lambda}\ln(\text{const}/\sqrt{|E-E_{\text{sep}}|})$ — a logarithm of the energy deficit, on top of the finite time around the rest of the loop. The $1/\lambda$ prefactor is why soft saddles produce dramatic lingering; the pendulum's version of this computation is worked in The Pendulum Phase Portrait and Separatrix.

Reversible but not conservative — what does such a system actually look like?

Take $\dot{x}=y-y^3$ , $\dot{y}=-x-y^2$ . Flip $t\to-t$ , $y\to-y$ : the first right side must be odd in $y$ (it is), the second even (it is) — reversible. Yet the system has no conserved quantity of the usual form and its flow is not area-preserving. Still the origin — a linear center — is a true center: a trajectory starting on the $x$ -axis near the origin arcs around and re-crosses the axis, and its mirror twin (the arc reflected in the axis, run backward) is also a solution; by uniqueness the two arcs seal into one closed orbit, exactly the third figure. Mechanics is full of doubly protected examples (the undamped pendulum is both conservative and reversible); the independent cases matter because reversibility survives some perturbations that destroy conservation, and vice versa.

Do these results survive in higher dimensions, where this chapter is eventually heading?

The theorems partially survive; the tidy portrait does not. Derivation A is dimension-blind: a conserved quantity forbids attractors in any dimension — a frictionless three-body problem can be endlessly complicated but never settles. What fails is the sufficiency of one conserved function: in four phase-space dimensions the set $E=\text{const}$ is a three-dimensional surface, and trajectories can wander over it chaotically instead of being pinned to closed curves. Whether further conserved quantities slice the energy surface down to tame tori is the integrability question, and the fate of those tori under perturbation is KAM theory — the far end of this chapter. The plane is the one arena where conservation alone finishes the job.

Conservative Systems and Hamiltonian Phase Flow

Where this lives today

The physical picture

What conservation means for a flow

Which tool answers which question

Limiting cases

Worked examples

1 · Full census of a double well (confidence builder)

2 · The quartic well that swings slower the smaller the swing (exam level)

The physicist

Going further

Before this

Where this leads